CONTEXT(CLONE)
Sign in

Grade protocol deviations across trial sites, with evidence

For: Clinical operations lead or trial quality manager at a sponsor or contract research organisation

Pattern: Map, verify, reduceNeeds scaleDesigned for 40 to 400 agents
Get early accessSign in to use

The pain today

Deviation logs and monitoring reports pile up across sites. The same deviation is graded differently from site to site, and patterns that point to a protocol design flaw surface late.

The ask

“I attached our protocol with amendments, the deviation classification plan, and the deviation logs and monitoring reports from all sites. For each deviation, tell me which protocol section it breaks and how our plan would grade it, and show me where sites graded the same thing differently.”

Plain words, as you would say it to a colleague. Edit it to fit your case before you send it.

What you attach or connect

  • Study protocol and amendments as text
  • Deviation classification plan
  • Deviation logs per site
  • Monitoring visit reports

The unit of work

One worker task per one logged deviation.

Why a swarm fits

Each deviation needs only its own log entry, the protocol section it touches and the grading rules. Entries are independent, and grading drift only shows when all of them are lined up.

Not for

A single-site study with a short log: one person reads it faster. It does not assess the effect of a deviation on subject safety or data integrity.

The decision tree

6 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

  1. Planner, while planning

    Scope checkYes or no, with a probability

    Before work starts on a unit

    Does this log entry describe a departure from a procedure the protocol requires, rather than an administrative note or a data query?

    Sees only: One deviation log entry

    Why: Keeps notes and queries out of the graded set before any worker is paid to read them.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenEscalate to a strong model
    • No: below 0.30thenSkip this unit

  2. Before workers, before a task runs

    Small worker or strong modelA choice among options

    Before a task runs

    Do the dates in this entry fall under one protocol version at this site, or span an amendment?

    Sees only: One log entry and the site's amendment approval dates

    Why: Sends only the entries that straddle an amendment to the expensive model.

    • One version in forcethenAccept
    • Spans an amendmentthenEscalate to a strong model
    • Dates missing from the entrythenMark unresolved

  3. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Does the quoted protocol passage require the specific visit window, assessment or criterion that the log entry says was not followed?

    Sees only: One log entry, the quoted protocol passage and its version date

    Why: Stops a deviation from being pinned to a protocol section that does not contain the requirement.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry

  4. Evidence checkYes or no, with a probability

    After a worker answers

    Does the quoted classification rule list this type of deviation under the grade the worker proposed?

    Sees only: The proposed grade, the quoted rule from the classification plan and the entry

    Why: Keeps proposed grades tied to the sponsor's own plan rather than to a model's sense of severity.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenMark unresolved
    • No: below 0.50thenReject and retry

  5. Reconciler, while merging

    Conflict checkA choice among options

    While reconciling

    Do these two entries from different sites describe the same kind of departure from the same protocol section?

    Sees only: Two log entries with their mapped section and site-assigned grades

    Why: Surfaces grading drift between sites without choosing which site was right.

    • Same kind, same gradethenAccept
    • Same kind, different gradesthenMark unresolved
    • Different kinds of departurethenContinue

  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does the entry mention eligibility criteria, consent, dosing of study treatment or an unreported adverse event?

    Sees only: One log entry with its proposed grade

    Why: Routes every subject-relevant deviation to the medical monitor and quality lead before it counts as graded.

    Accountable: The medical monitor and the sponsor's quality lead own the final grade and any decision to report a serious breach.

    • Yes: 0.40 or higherthenAsk a person
    • Unsure: 0.15 up to 0.40thenAsk a person
    • No: below 0.15thenAccept

The fleet: who does what

Model tiers by role, not brands: you choose the models. Strong reasoning models plan and reconcile, small fast models do the wide work, and the judge is a decision model from a different family, so it does not share the workers' blind spots.

  1. Planner

    A strong reasoning model indexes the protocol by section and amendment date and routes each deviation to its slice.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from an open-weight family map one deviation to a protocol section and propose a grade with quotes.

    Designed for 40 to 400 agents, one worker task per one logged deviation. Each worker receives only its own unit.

    Decisions here:2. Small worker or strong model

  3. Judge, from a different model family

    A decision model from a different family checks that the quoted protocol text and amendment version support the grade.

    Decisions here:3. Evidence check4. Evidence check

  4. Reconciler

    A strong reasoning model groups deviations by section and site and lists inconsistent grading without resolving it.

    Decisions here:5. Conflict check

  5. Accountable person

    The medical monitor and the sponsor's quality lead own the final grade and any decision to report a serious breach.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Every proposed grade quotes the protocol section and the classification rule used
  • The amendment in force on the deviation date is checked
  • Deviations the judge cannot support are returned ungraded, not guessed

What comes back

  • Deviation table with protocol section, proposed grade and quotes
  • Same-type deviations graded differently across sites
  • Protocol sections that attract repeated deviations
  • Ungraded items with the reason

What to measure

  • Agreement with the quality manager's grades on a sample
  • Share of deviations returned ungraded
  • Reviewer time per deviation
  • Cost per deviation

Names of measures only. No result is claimed for this template.

Templates open in the workspace chat with the ask filled in. Nothing runs until you send it.

Get early accessSign in to use

More in Healthcare and life sciences

Check adverse-event narratives against their case data

For: Pharmacovigilance lead or safety writer preparing case narratives for a study report or aggregate report

Hundreds of case narratives are written from line listings.

Pattern: Cross-examinationNeeds scale5 decisionsDesigned for 30 to 300 agents